Scarfski/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2026-09-13 05:48:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

bumping "down" CUDA version

Browse Source
This commit is contained in:
Marcos Slomp
2026-06-03 11:05:06 -07:00
parent cc76d0f60e
commit 224ff6d0e8
2 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
manual/tracy.tex
View File

@@ -1773,7 +1773,7 @@ Similar to Vulkan and OpenGL, you also need to periodically collect the OpenCL e
\subsubsection{CUDA}
CUDA support is enabled by including the \texttt{public/tracy/TracyCUDA.hpp} header file. To use it, make sure you have the NVIDIA CUDA Toolkit v12.8 (or later) installed, and that the NVIDIA CUPTI library is available in the toolkit (located at \texttt{CUDA\_INSTALLATION\_PATH/extras/CUPTI}).
CUDA support is enabled by including the \texttt{public/tracy/TracyCUDA.hpp} header file. To use it, make sure you have the NVIDIA CUDA Toolkit v12.4 (or later) installed, and that the NVIDIA CUPTI library is available in the toolkit (located at \texttt{CUDA\_INSTALLATION\_PATH/extras/CUPTI}).
Tracing CUDA requires the creation of a Tracy CUDA context using the macro \texttt{TracyCUDAContext()}, which returns an instance of a \texttt{tracy::CUDACtx} object. TracyCUDA allows only a single \texttt{tracy::CUDACtx} object at any given time. Subsequent calls to \texttt{TracyCUDAContext()} will return the same reference-counted object. There is no need for clients to instantiate multiple \texttt{tracy::CUDACtx} objects, as a single context is capable of instrumenting all CUDA contexts and streams.

11
public/tracy/TracyCUDA.hpp
View File

@@ -19,8 +19,8 @@ using CUDACtx = std::nullptr_t;
#else
#include <cupti.h>
#if CUDA_VERSION < 12080
#error "CUDA v12.8 (or later) is required by TracyCUDA.hpp"
#if CUDA_VERSION < 12040
#error "CUDA v12.4 (or later) is required by TracyCUDA.hpp"
#endif
#include <cassert>
@@ -1215,8 +1215,15 @@ namespace tracy
{
// NOTE(marcos): a byproduct of CUPTI_ACTIVITY_KIND_SYNCHRONIZATION
// (I think this is related to cudaEvent*() API calls)
#if CUDA_VERSION < 12080
// prior to CUDA v12.8
CUpti_ActivityCudaEvent* event = (CUpti_ActivityCudaEvent*)record;
UNREFERENCED(event);
#else
// starting from CUDA v12.8
CUpti_ActivityCudaEvent2* event = (CUpti_ActivityCudaEvent2*)record;
UNREFERENCED(event);
#endif
break;
}
default: